Skip to main content

Verification infrastructure for home lending

Your controls assume the worker is human.

AI changes the worker — not the obligation to prove what happened. wetink seals every judgment it records — who judged, against which version of which rule, citing which page of which document, at what cost — into a record no one, including us, can quietly change. The proof trail is the product.

Request an architecture briefingWhy now

Sealed judgment record

sealed
SPECIMEN RECORDschema enforcedvalues illustrative
Who judged checker_04
Rule version FNMA / 2026.08.06
Source W2_2025.pdf / p.1
Maker worker_13
Checker reviewer_04
Prior entry 81fc…
Cost $0.0184
append-only — prior entry carried forward

The evidence

The evidence — passing proofs, not a demo

123/123 — PASS

47 seam proofs + 24 reviewer witness/manifest proofs + 40 headless-loop proofs + 12 migration-runner proofs, as the wetink repository's own suite counts state them.

2

distinct actors required to seal

Maker and checker are separate database roles with no membership between them; a maker physically cannot seal its own work.

0

accuracy percentages quoted on this site

None exists to quote, so none appears. That is the register this product stands on.

Product status

What runs today

wetink draws a hard line between what is built and proven now and what is roadmap, and keeps that changing detail in one authoritative place. The sealing and enforcement kernel exists and holds every judgment behind 123 passing proofs; the AI judgment layer that would evaluate a claim does not exist yet, and there are no customer deployments today.

See what runs today, in full — Architecture

Enforcement

The maker physically cannot seal its own work.

The schema separates the actor who does the work from the actor who approves it — no membership between the two roles. When the same actor tries to do both, the write is refused before it reaches the chain, not flagged after the fact.

refusal readout 
finding recorded      reviewer A    insufficient evidence
superseding pass      reviewer A    not accepted — same actor as the original finding
superseding pass      reviewer A    not accepted — same actor as the original finding
seal                                blocked until a distinct reviewer resolves the finding
From the reference implementation this enforcement pattern was proven in — a separate, predecessor platform. Not wetink product or customer history.

The mechanism

What must happen before a judgment becomes authoritative

FIG. 02 — The judgment path

The full lifecycle, with what the schema refuses at each step ·  One decision, walked — the worked example

Applications

Start with post-close QC.

Post-close QC is where the same judgment shape first proves itself: full-population reperformance instead of the 10% sample every lender already runs. It is the wedge, not the whole architecture.

WORKED EXAMPLEPOST-CLOSE QC

The worked example: full-population reperformance instead of the 10% sample.

OriginationUnderwritingClosingPost-close QCServicingAudit / disclosure

Interface preview

The console a sealed record gets

A review queue, the sealed record beside it, and the schema's refusals rendered as first-class states — a design preview, not a product capture.

wetink — review queuePrerelease interface design
Review queue (illustrative)
FileChecksStatus
Purchase file12/12sealed
Refinance file9/12awaiting distinct checker
HELOC fileEXTRACTION_UNANCHOREDrefused

sealed record

Who judged maker_07 + checker_04 (distinct)
Rule version v2026.08.06
Anchor W2_2025.pdf · p.1

prior-entry hash carried forward — append-only

Illustrative schema states from the sealing architecture that runs today; the review layer that would fill this queue does not yet exist — see what runs today.

The economics

The control system samples because humans are expensive.

Fannie Mae's Selling Guide D1-3-01 sets the floor: a minimum of 10% of the loans a lender originates or acquires, selected randomly, for post-closing QC review. The incumbent category manages that sample. wetink built its architecture for the other90% too.

A design basis, not a delivered customer result — the review layer that will produce it does not yet exist. Source for the 10% floor: Fannie Mae Selling Guide D1-3-01.

First commercial deployment

Selecting one lending partner.

We are validating the first commercial deployment with lending organizations where machine-assisted work creates a new governance problem — a paid, narrowly scoped founding pilot with defined deliverables, never free product discovery.

  • Paid
  • Production
  • Narrowly scoped
  • Defined deliverables
  • One selected workflow

See the engagement structure →

Dated pieces

Writing

Regulatory notePublished

Disclosure on demand: what LL-2026-04 actually requires

Fannie Mae's AI governance framework took effect on 2026-08-06. What it obligates, and what an answer built to be checked looks like.

By wetink

All writing

Thirty minutes on the architecture. No product theater — there is no product to demo yet, and we say so.

  1. 01The record
  2. 02The enforcement
  3. 03The fit
Request an architecture briefingFounding partner →

Or write directly: hello@wetink.ai